Yesterday we wrote about the hosts flexing. Canada traveled 0.9km to BC Place. The model said 67%. Canada scored six.
June 18: the scorecard
| Match | Result | P(result) | Brier |
|---|---|---|---|
| Mexico vs South Korea | 1-0 | 45.3% | 0.448 |
| South Africa vs Czech Republic | 1-1 | 24.7% | 0.938 |
| Canada vs Qatar | 6-0 | 67.4% | 0.167 |
| Bosnia vs Switzerland | 1-4 | 68.9% | 0.153 |
Mean Brier: 0.426. Uniform baseline: 0.667. The model's best day of the tournament by a wide margin.
Canada 6-0: the hosts deliver
The model said 67.4%. That was the widest favourite on the card. The travel gap (0.9km vs 11,697km) was the second-largest in the entire group stage. Everything pointed to a Canada win.
Nobody predicted six-nil. The model saw the direction, not the magnitude. When we previewed the match, we wrote "Qatar will make this difficult. Their tactical fingerprint is low-block. They will absorb, frustrate, and wait." Qatar absorbed nothing. Canada dismantled Lopetegui's defensive shape and delivered the most emphatic win of the tournament.
Brier: 0.167. The model's third-best single-match score. Jonathan David and BC Place delivered on everything the preview promised and more.
Switzerland 4-1: the response
After the Qatar draw on Matchday 1, which produced a Brier of 1.108 (the worst single-match penalty of the tournament), Switzerland's credibility was on the line. We wrote "that number deserves a question mark."
Switzerland answered. Four goals against Bosnia in Los Angeles. Brier: 0.153, the best single-match score of the day and the model's second-best of the tournament. The question mark is gone.
Edin Dzeko, 40 years old, 148 caps, watched from the pitch as his second World Cup slipped away. Bosnia's elimination is not yet mathematically confirmed, but the path from here requires results they have not shown.
South Africa 1-1: draw number ten
Both teams lost on Matchday 1. Both faced elimination. Neither blinked.
The model had Czech Republic at 58.7%. The draw at 24.7%. Through 28 matches, the tournament's draw rate sits at 35.7%. The model's average predicted draw probability across all 28 matches remains around 24%. That gap has not closed.
Brier: 0.938. A heavy miss. The model's draw problem continues: when it assigns low probability to a draw, draws keep arriving.
Ronwen Williams kept South Africa alive. Czech Republic created chances but could not find the winner. Both teams now have one point from two matches. Their final group match will decide everything.
Mexico 1-0: the quiet one
The model called this a three-way split: 45.3% Mexico, 27.4% draw, 27.3% South Korea. The closest thing to a coin flip on the card. Mexico won quietly, professionally, at home, 473km from Mexico City.
Son Heung-min flew 11,732km and shifted 15 timezone hours. Mexico's home advantage was always the story. The model's 45.3% was cautious, and the result validated caution over conviction: Mexico won, but a draw or a South Korea win would not have surprised anyone.
Brier: 0.448. A solid score for a match the model essentially treated as open.
The model's best day
Mean Brier: 0.426. For context, the previous best day was June 16 (0.197, but that was a four-match card where every heavy favourite delivered). June 18 included a genuine coin flip (Mexico vs South Korea) and an upset-draw (South Africa vs Czech Republic), and the model still produced its best overall performance.
Three of four outcomes went the model's way. The only significant miss was the South Africa draw. When the card includes both clear favourites and open matches, the model needs the favourites to land. They did.
Calibration through 28 matches
Running mean Brier: 0.627. Uniform baseline: 0.667. The model continues to outperform random guessing, but the margin is narrow.
Ten draws from 28 matches. 35.7%. The model's average predicted draw probability: roughly 24%. The tournament keeps producing more draws than any model anticipated. Whether that is a feature of the 48-team format, the expanded geography (teams crossing 10,000+ kilometres to play), or something else entirely, the pattern is real and the model has not adapted.
June 19: the hosts are underdogs
Groups C and D return for Matchday 2. The card features the tournament's biggest mismatch, its most counter-intuitive model call, and two elimination matches.
| Match | Group | Venue | Model H/D/A | Travel gap |
|---|---|---|---|---|
| USA vs Australia | D | Seattle | 31.3 / 28.2 / 40.5 | 10,276km (USA closer) |
| Morocco vs Scotland | C | Boston | 48.1 / 28.4 / 23.5 | 782km (roughly equal) |
| Brazil vs Haiti | C | Philadelphia | 90.4 / 8.7 / 0.9 | 4,139km (HAI closer) |
| Paraguay vs Turkey | D | San Francisco | 31.4 / 27.8 / 40.9 | 1,733km (PAR closer) |
USA vs Australia: the model picks the visitors
The model rates Australia at 40.5%. The USA at 31.3%. Draw at 28.2%.
Read that again. In the USA's own World Cup, at Lumen Field in Seattle, the model says the visitors are more likely to win.
Australia flew 13,065 kilometres to get here. A 17-hour timezone shift. The longest single-match travel in today's card. The USA drove 2,789 kilometres from their base with a three-hour shift. Home crowd. Home conditions. The structural advantage could not be more one-sided.
The model sees it differently. Australia's Elo sits at 1783, 62 points above the USA's 1721. Both teams won on Matchday 1, but the model's pre-tournament priors weigh heavily. Australia beat Turkey 2-0 in their opener. The USA beat Paraguay 4-1, a more emphatic margin, but against a slightly lower-rated opponent.
Tony Popovic coaches Australia. Mauricio Pochettino coaches the USA. Two managers appointed in 2024, both building squads for this moment. Pochettino's experience at Tottenham, PSG, and Chelsea dwarfs Popovic's resume on paper, but Popovic's Australia have been efficient and disciplined.
Christian Pulisic (4.2% match scoring probability) leads the US attack. Folarin Balogun (8.9%, USA's penalty taker) carries the higher individual threat. For Australia, Brandon Borrello (3.5%) and Mitch Duke (2.7%) are modest by comparison. The scorer probability gap favours the USA, even as the match probability does not.
Australia are "transition-heavy" (directness 7.2, 46 long balls per 90). The USA are "balanced" with no dominant tactical identity. Australia's approach, the long direct ball into space, is designed for exactly this scenario: absorb pressure in a hostile environment, then strike on the counter.
Group D standings: USA 3 points, Australia 3 points, Paraguay 0, Turkey 0. The winner here takes a massive step toward the knockout rounds. A draw keeps both alive but opens the door for Paraguay or Turkey to close the gap.
This is the match that defines whether the model's travel adjustments are strong enough. If Australia win in Seattle, the Elo signal is stronger than home advantage. If the USA win, the model underpriced what it means to play at home.
Brazil vs Haiti: the numbers
90.4% for Brazil. 0.9% for Haiti. 8.7% draw.
The Elo gap is 452 points (Brazil 1984, Haiti 1532). This is one of the largest gaps in the tournament. Haiti's 1532 rating is the lowest of any qualified team.
But Brazil drew Morocco 1-1 on Matchday 1. Carlo Ancelotti's side were supposed to cruise through Group C. Morocco's low-possession counter-attack frustrated Brazil's high press (PPDA 17.1, one of the most intense in the tournament). Ancelotti needs three points here. Anything else raises genuine questions about Brazil's group-stage passage.
Haiti are here for the first time. They lost 0-1 to Scotland on Matchday 1. They traveled 2,344 kilometres to Philadelphia, actually closer than Brazil's 6,482km. A small advantage in a match where every small advantage is buried under a 452-point Elo gap.
Raphinha (13.7% match scoring probability, Brazil's penalty taker) is the most likely scorer on today's card. Gabriel Jesus (6.7%) provides the secondary threat. Against Haiti's "balanced" tactical profile (a polite way of saying the data is too thin to characterize their style), Brazil's press should dominate.
The model says 90.4%. If this is the match where 90% does not deliver, it would be one of the biggest shocks in World Cup history. The honest assessment: this is the closest thing to a certainty on the schedule.
Morocco vs Scotland: the defensive wall vs the counter-attack
Morocco at 48.1%. Scotland at 23.5%. Draw at 28.4%.
Morocco drew Brazil. Scotland beat Haiti. Both enter Matchday 2 with something to build on, though the quality of opponent faced on Matchday 1 tells very different stories.
The tactical matchup is unusual. Morocco are "counter-attackers" (low possession 0.46, vertical buildup). Scotland are "low-block" (PPDA 26.0, low possession). Two teams that do not want the ball. Two teams that wait for the opponent to commit. When both sides adopt the same defensive posture, the result tends toward stalemate.
Sofyan Amrabat (6.2%, Morocco's penalty taker) against Che Adams (3.4%). Achraf Hakimi provides width from right-back. Andrew Robertson provides it from left-back for Scotland. Two full-backs who will determine the shape of the match, because the central channels may be locked shut.
Steve Clarke has managed Scotland since 2019, the longest tenure among today's four managers. His record is built on exactly this profile: defensive discipline, low block, wait for mistakes. Morocco's Mohamed Ouahbi, appointed in 2026, is the newest manager on the card.
This match has the highest draw probability on the card (28.4%) and the tactical profile to justify it. Two defensive sides, similar travel (Morocco 5,682km, Scotland 4,900km), no dominant attacking threat. The model assigns a Morocco lean, but this screams draw.
Paraguay vs Turkey: both far from home
Paraguay at 31.4%. Turkey at 40.9%. Draw at 27.8%.
Both lost on Matchday 1. Paraguay fell 1-4 to the USA. Turkey lost 0-2 to Australia. Both face elimination. The loser's World Cup is almost certainly over.
Paraguay traveled 9,477 kilometres to Levi's Stadium in San Francisco. Turkey traveled 11,209 kilometres. Both face a four-hour-plus timezone shift. Neither has anything resembling home advantage. This is as neutral as a World Cup match gets.
The model slightly favours Turkey. Their Elo (1902) sits 69 points above Paraguay (1833). Hakan Calhanoglu (6.9%, Turkey's penalty taker) and Kenan Yildiz (3.2%) lead the Turkish attack. Antonio Sanabria (7.9%, Paraguay's penalty taker) and Julio Enciso (4.7%) counter for Paraguay. The individual scoring threats are more evenly matched than the model's headline number suggests.
Turkey are "pragmatic" (mid-band on every axis). Paraguay are "balanced." Two teams without a dominant tactical identity, both facing the kind of desperate pressure that Matchday 2 elimination matches produce. Gustavo Alfaro (Argentina) coaches Paraguay. Vincenzo Montella (Italy) coaches Turkey. Two foreign coaches, both appointed in 2023-2024, both staring at an early exit.
The draw question, continued
Ten draws from 28 matches: 35.7%. The model's average predicted draw probability across today's four matches: 25.3%. The model implies one draw from today's card. The tournament's observed rate implies at least one, possibly two.
Morocco vs Scotland (28.4%) has the highest draw probability and the defensive tactical profile to match it. USA vs Australia (28.2%) is close to a three-way split, the kind of match that draws more often than the model's headline favourite suggests.
The model assigns draw probabilities between 8.7% (Brazil vs Haiti) and 28.4% (Morocco vs Scotland). The range is wide. But the model has shown no ability to discriminate between matches that draw and matches that do not. Its average predicted draw probability on actual draws (roughly 24%) is indistinguishable from its average on non-draws.
If the pattern holds, expect at least one draw that the model did not favour.
All probabilities are frozen pre-match model outputs, locked before kickoff. The model publishes probabilities, not recommendations. Methodology: /docs/methodology/. Track record: /accuracy/. Full Terms of Use.